JP2958649B2 - Composer and composition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a composer and a composition method for automatically generating a melody.

[Background] There are many phenomena (time-series phenomena) in the natural world and human activities that have a characteristic correlation (the property that current events depend on past events). In terms of music, many existing songs have a note spectrum with a power spectrum of 1 /
It is known that it has a frequency characteristic close to f (where f is the fluctuation frequency). The frequency characteristic referred to here is a characteristic relating to a mode in which a time series fluctuates, and is not an attribute of a time series element itself (for example, a frequency indicating each pitch in a pitch series). For example, a completely random pitch sequence (frequency sequence) forms white noise, and its fluctuation frequency characteristics have a constant gain (1 / f ⁰ = 1) independent of frequency.

Some songs deviate from the fluctuation frequency characteristic of 1 / f, but it can be recognized that these songs are perceptually different from the music group having the characteristic of 1 / f.

The fluctuation frequency characteristic of a music is a characteristic of the music that is determined by analysis (spectral analysis). Conversely, a music is synthesized (composition) from information specifying such a characteristic or using a fluctuation frequency characteristic time-series generator. The attempt to do
f Automatic fluctuation composer “Shaharazard” using fluctuations (Introduction to Microcomputer Sound, Kosaido Publishing, November 10, 1981, first edition, pp. 144-157, Masahiko Narita) and “Music Toy Program” (bit separate volume, No. 217-225, Kuniaki Tsuboi).

In these documents, a fluctuation generator that approximately generates 1 / f fluctuation is realized by a white noise source (implemented by a pseudo-random number algorithm of a program) and a digital filter that combines multiple stages of an integration filter and a differentiation filter (also implemented by a program) And a random number sequence from a white noise source is passed through this).

Unfortunately, in the case of this conventional technique, it is necessary to redesign the digital filter in order to change the characteristics of the fluctuation generator, and a large number of different digital filters are required to compose music having various fluctuation frequency characteristics. It is necessary, and there is a problem that it is expensive and the configuration becomes complicated. In addition, since each value of the time series output from the fluctuation generator is regarded as the frequency of the pitch and is allocated to the scale tone as it is, it is not possible to control the pitch range, and the possibility of an unnatural scale tone series is denied. Can not. For example, for many songs, the pitch at the climax is a specific pitch that can often be expected from the melody flow and gives a favorable pitch feeling, but the fluctuation generator itself compensates for the fluctuation characteristics of its output time series However, it does not guarantee a pitch value that is preferable.

In other words, if the pitch sequence of the melody of an existing music is linearly transformed in the pitch direction (for example, a transformation in which the highest note rises or falls by a semitone), the fluctuation characteristics are preserved, Obtained by an unnatural melody on the pitch sensation (especially on the pitch sensation which is a relative pitch).

[Object of the Invention] Accordingly, an object of the present invention is to provide a composer capable of generating a melody having various fluctuation frequency characteristics while having a relatively simple configuration and allowing a user to easily select a desired fluctuation frequency characteristic. Is to provide a composition method.

It is a further object of the present invention to provide a composer capable of relatively easily generating a melody having a fluctuation frequency characteristic and suitable for a pitch sense.

According to one aspect of the present invention, a random number generating means for generating a time series of random numbers, a variable setting means for setting a selection value of a variable number, Coefficient generating means for generating a linear conversion coefficient for a random number time series in accordance with the set selection value; and linearly transforming the random number time series with the linear conversion coefficient, so that the fluctuation frequency varies for each variable number value. Linear conversion means for selectively generating a time series having a fluctuation frequency characteristic with respect to the variable selection value from the time series having the characteristic; and a similar fluctuation based on the time series generated by the linear conversion means. A melody generating means for generating a melody having a frequency characteristic.

In the case of this configuration, since the fluctuation frequency characteristics of the time series generated by the linear conversion means are different for each linear conversion coefficient, various fluctuation frequency characteristics can be obtained for the common linear conversion means only by changing a parameter called a linear conversion coefficient. Can generate a time series with That is, it is possible to generate a melody having various fluctuation characteristics without changing the form of the operation performed using the linear conversion coefficient and the time series of random numbers (therefore, the configuration of the functional linear conversion means). Further, the coefficient generation means determines a linear conversion coefficient corresponding to the variable number of selected values. As a result, the variable number of selected values functions as parameters that selectively specify the fluctuation characteristics. Therefore,
The user can easily obtain a melody having desired fluctuation characteristics.

As one configuration example, 1 / f ⁿ (where f is the fluctuation frequency, n
, A random number generating means for generating a time series of random numbers, and a characteristic corresponding to a selected value of the order n. Characteristic setting means for variably setting the designated data, coefficient generating means for generating a linear conversion coefficient for the time series of the random numbers in accordance with the set property designation data, and linearizing the time series of the random numbers with the linear transformation coefficients And a linear conversion means for generating a time series having a 1 / f ⁿ approximate frequency characteristic having the selected value as an order.

In this configuration, the time series output from the random number generation means has a constant gain whose fluctuation frequency characteristics do not depend on the fluctuation frequency. On the other hand, the linear conversion means functions as a linear filter, and assigns a fluctuation frequency characteristic corresponding to the linear conversion coefficient to the random number time series from the random number generation means.

Instead of the random number generating means, an original time series generating (giving) means for generating or adding an original time series having a predetermined fluctuation frequency characteristic may be used. For example, the original time series assigning means can be constituted by a means for sampling an appropriate time series generated by the nature of nature or artificially. In this case, the fluctuation frequency characteristic of the time series output from the linear conversion means is a combination of the fluctuation frequency characteristic of the original time series and the frequency response characteristic of the linear conversion means.

In a preferred configuration example, the linear conversion coefficient is given by a polynomial of specific designation data designated by the characteristic setting means. (Where a _r is the r-th (r ≧ 1) linear transformation coefficient, n is specific designation data), and a ₀ is given by a ₀ = 1. In the case of the above equation, the frequency characteristic of the linear conversion means is 1 /
is approximated by f ⁿ .

Further, according to the present invention, in a composition method for generating melody data based on an output of a time series having a fluctuation frequency characteristic approximated to 1 / f ⁿ (where f is a fluctuation frequency and n is an order), A random number time series is generated by using a random number generation means, the characteristic designation data corresponding to the selected value of the order n is variably set by the input means, and the time series of the random number is linearly set by the coefficient generation means. A conversion coefficient is generated according to the set characteristic specification data, and the time series of the random numbers is linearly converted by the linear conversion means using linear conversion means, and the 1 / f ⁿ approximation fluctuation having the selected value as an order is used. A composition method is provided, comprising generating a time series having frequency characteristics.

According to this configuration, the melody data having the specified fluctuation frequency characteristic can be obtained only by specifying the desired characteristic specification data from the input unit.

Further, according to the present invention, a time series generating means for generating a time series having a fluctuation frequency characteristic approximated by 1 / f ⁿ (where f is a fluctuation frequency, and n is an order), A composer is provided, comprising: a range setting unit for setting to be variable; and a time series range conversion unit for converting the generated time series into a time series having the set range.

According to this configuration, it is possible to easily obtain a melody that is natural in terms of a sense of pitch or a melody having a sound range that the user likes.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows the entire configuration of an automatic composer in which the principle of the present invention is mainly applied to generation of a melody pitch train.
The present invention is also applicable to control of generation of a melody note length sequence.

The random number generating means 1 has an appropriate value range, for example, as shown in FIG.
Generate a time series {x _i } of random numbers x _i in the range from 1 to 1. The random number generating means 1 includes an arbitrary well-known random number generator (for example, a random number generator based on a congruential method, a random number generator based on a shuffle method, a random number generator based on a maximum periodic sequence (M sequence), and the like). Can be used. Each element of the random number time series {x _i } from the random number generation means 1 is input to the linear conversion means 2, where the linear conversion coefficients a ₀ , a ₁ , a a given from the degree designation means 3 through the coefficient generation means 4 ₂ ......
_{aN, that} is, linearly transformed by { _ar }. Here, the order designating means 3 is a means for variably instructing the frequency response characteristic (linear conversion characteristic) of the linear conversion means 2, and corresponding to the order n as the characteristic designation data, the order of the frequency response of the linear conversion means 2. The spectral characteristics are given by approximately 1 / f ⁿ (where f is the frequency).

In order to ensure the correspondence between the degree specification data n and the frequency response characteristic of the linear conversion means 2, the coefficient generation means calculates a ₀ = 1, Generate each linear coefficient a _{r represented} by

Here, the polynomial, the frequency response characteristic of the linear transformation means 2 using linear coefficients a _r, therefore, the fluctuation frequency characteristic approximately 1 / f ⁿ of the time series after passing through the random number x _i in linear transformation unit 2 Let's explain that it has the characteristic of

Now, assuming that the random number from the random number generation means 1 is an ideal random number, its power spectrum is constant regardless of the frequency. This can be expressed fluctuation characteristics are as f ^0. If the ideal random number sequence having this fluctuation characteristic f ⁰ is differentiated once, its fluctuation characteristic is f ² , if it is differentiated twice, it is f ⁴ , and if it is differentiated n times, it is f ²ⁿ , and conversely, it is integrated once. The fluctuation characteristic is 1 /
f ² , if integrated twice, 1 / f ⁴ , and if integrated n times, 1 / f ⁴
f ²ⁿ .

Although the above is a discussion on a continuous time series, the same applies to a discrete time series.

The differential in the continuous time series corresponds to the difference in the discrete time series area. Therefore, (Where Δt is a discretized time interval) Therefore, assuming Δt = 1, Is Corresponding to However, when r = 0, the coefficient of x _m is 1. Here, if -n is used instead of n, an equation corresponding to n-times integration (and thus its fluctuation characteristic 1 / ^f2n ) is obtained. That is, (Where the sum is truncated to the term N). Here, if n is replaced by a real number n / 2, And the coefficient of x _m matches the equation shown first. Y _m above
In the equation, the number of terms of the sum (the number of x _{m to be} used) is truncated to a finite number.

Assuming that the number of random numbers used in one linear exchange is N and the discrete time interval is Δt, the value of N and Δt is 1 / f ⁿ
The lower limit of the frequency band having the shape fluctuation characteristic is determined, and the upper limit is the Nyquist frequency.

Although the above is the theoretical explanation, FIGS. 4 and 5 show examples of characteristics in the case of actually calculating. FIG. 4 shows that when the order n is 1, the random number time series {x _i } is
Is the average value of 100 times the fluctuation frequency characteristic (power spectrum) of the time series of 256 samples obtained by passing through an N = 256-stage FIR digital filter (see FIG. 3),
FIG. 5 shows similar average values of fluctuation frequency characteristics when 1.5 is selected as the order n. In both figures, the horizontal axis represents frequency (the Nyquist frequency is indicated by 128), and the vertical axis represents gain. As shown in the figure, for the order n = 1, an approximately 1 / f type fluctuation frequency characteristic is obtained, and n = 1.5
, A fluctuation frequency characteristic of about 1 / f ^1.5 was obtained.

FIG. 2 shows a configuration example of the coefficient generating means.
Block 20 is implemented with a translation table or calculator such as a ROM. In the case of a table configuration, the designated data n of the order is used as the first partial address, and the count from the coefficient counting stepping circuit 21 (a counter that sequentially outputs numerical values from 0 to N in response to the start signal) is performed. A table storing the r-th linear transformation coefficient data a _r for the order n may be used in a storage location where r is the second partial address. In the case of a calculator, a required polynomial (for example, the above-described polynomial) may be calculated using the coefficient number r and the order n as arguments.

FIG. 3 shows a configuration in which the linear conversion means is constituted by an FIR type digital filter. N cascaded delay elements 30-1 to 30-N (implemented by shift registers or memories) and (N + 1) coefficient multipliers 31-0 to 31
−N and the accumulator 32 are used. The random number time series {x _i } from the random number generation means 1 is added to the FIR digital filter as an input signal. Random number signal x _i applied sequentially transmitted to the right direction shown the chain 30-1 to 30-N of delay elements.
The content of the immediately preceding delay element 30-1 is _xn-1 , the content of the immediately preceding delay element 30-2 is _xn-2 , and so on, and the content of the last delay element 30-N is _xnN. A new random number x _n
, The N random number data x _n , x _n-1 ... X _nN
To _{each, a 0, a 1 ...... a} N weights each of the coefficient multipliers 31-0,31-1, attached at ...... 31-N, the sum is taken at accumulator 32, the new linear transformation data y _n are generated. Here, the linear conversion data y _n is _{y n = a 0 x n +} a 1 x n-1 +
...... + given by a _N x _nN. The coefficients a ₀ , a ₁ ,..., _An are data from the coefficient generating means 4.

The linear transformation time series {y _i } generated in this way is stored in the memory means 5 as shown in FIG. As is clear from the above description, this time series {y _i } has a fluctuation frequency characteristic approximated by 1 / f ⁿ with respect to the order n specified by the order specifying means 3. Subsequent processing relates to conversion of a format in which the fluctuation frequency characteristic is basically preserved, and as a result, the pitch time series of the generated melody also has the same fluctuation frequency characteristic. In other words, the time series stored in the memory means 5 determines the fluctuation frequency characteristics of the pitch time series of the target melody, and this information is the basis of the melody.

Blocks 6, 7, and 8 which are processing elements for the time series {y _j } of the desired fluctuation frequency characteristic are obtained by scaling the value of each element y _j in the time series {y _j } when the vibration having the designated sound range is performed. This is for generating a sequence. For details,
The maximum / minimum value extracting means 6 extracts an element y _max having the maximum extraction and an element y _min having the minimum value in the time series {y _j }. On the other hand, data f _max and f _min that specify the range of the frequency series {z _j } output from the frequency series calculating means 7 are input. In other words, the time series element with the maximum value y _max is regarded as the element having the frequency of f _max , and the time series element with the minimum value y _min is
It is regarded as an element having a frequency of f _min . Therefore, the frequency series calculating means 7 calculates the time series {y _j } from the memory means 5 To generate a frequency sequence {z _j }.

These means 6, 7, 8 take into account the human sense of pitch. That is, the relative pitch in music needs to be a value that is considered musically favorable. The pitches of particular concern include the highest note and the lowest note. This is because the highest and lowest sounds are often more worrisome than the sounds of the cognitive upper ground. Therefore, by adopting the configuration of the means 6, 7, 8 as shown in the drawing, at least the highest tone and the lowest tone can be set to sounds of appropriate pitches preferred by the user. It is possible to obtain a melody with a good efficiency relatively efficiently.

A block 9 for processing the output of the frequency series calculating means 7;
10 is for establishing a pitch between discrete pitches. That is, music generally has a plurality of discrete pitch sounds recognized as scales. It is thought that the scale sound is related to the psychological stability of music. Also, various scales are known, and the type of the scale is one element of the musical style. Therefore, the scale designating means 9 designates the scale composition of the melody to be composed, and the scale assigning means 10 assigns each element of the frequency series {z _j } to the scale sound having the closest frequency according to the specified scale. Performs the discretization process assigned to. Thus, the pitch sequence of the melody to be composed is completed (although editing may be performed through a display or the like after this).

The note length determining means 11 determines the length of each note of the melody, and completes the note length sequence of the melody. The pitch sequence of the melody generated by the scale assigning means 10 and the pitch sequence of the melody generated by the pitch determining means are stored in the memory means 12 as melody data.

The main processing elements and storage elements shown in Fig. 1 can be realized by a DSP (Digital Signal Processor) for high-speed processing, but can be realized by a general-purpose microcomputer if there is no problem in processing time. it can.

FIG. 6 shows an example of a configuration in which the microcomputer shown in FIG. 1 is realized by a microcomputer. In the example shown in the figure, the function of the above-described automatic musical composition is incorporated in a keyboard-type electronic musical instrument.

In FIG. 6, a CPU 100 operates according to a program stored in a ROM 101 and controls each circuit element. The keyboard 102 is used for playing a musical instrument. The input device 103 has order n
This is used for inputting the designation of a pitch, the designation of a range, the designation of a scale, and the like, and realizes the functions of the degree designation means 3, the range determination means 8, and the scale designation means 9 in FIG. Work RAM is CPU
It is used as 100 working memories and temporarily stores input data, linear transformation coefficients, and the like. The song memory (RAM) 105 stores the composed melody. The musical score output device 106 includes a display device such as a CRT and a printer, and is used for outputting a musical score. Edit the melody data in the song memory 105 once it has been automatically composed using the CRT and the input device 103 if desired (selection of melody, chording, tempo setting, etc.)
It can be performed. Tone generator 107 is CPU100
Electronically generates a tone signal under the control of the audio system 108, and the audio system 108 converts the tone signal into an audio signal.
The speaker 109 converts an audio signal into an audio signal.
By using these elements 107, 108, and 109, melody data and the like in the music memory 105 can be automatically played.

FIG. 7 is a flowchart of the automatic music composition process of the CPU 100 of FIG. At 7-1, the order n specifying the fluctuation characteristic 1 / f ⁿ of the melody to be composed, the range of the melody, and the scale configuration of the melody are input from the input device 103. Next, 7
A set of linear coefficients {a _r } from the order n related to -2
(R = 0, 1,... N) are generated in the manner described above. Next, in a loop of 7-3, 7-4, and 7-5, random numbers x _i
While generating (−1 ≦ x _i ≦ 1) (7-3), the process of performing (7-4) the linear transformation on the random number sequence by a set of linear coefficients is repeated a required number of times to correspond to the designated order n. Do
Generate a time series {y _j } with 1 / f ⁿ fluctuation characteristics. Next, the maximum value y _max and the minimum value y _min in the time series {y _j }
Is extracted (7-6), and using the designated range (f _{min to} f _max ), , A normalized (range-limited) frequency sequence {z _j } is generated (7-7). Further, each frequency data of the frequency series {z _j } is assigned to the pitch of the scale note in accordance with 7-8 according to the designated scale, thereby obtaining a pitch sequence of the composition melody.
In step 7-9, the note length of each note of the composition melody is determined to generate a note length sequence of the composition melody. Finally, the melody relating to the composition is automatically performed via the tone generator 107, the audio system 108, and the speaker 109 (7-10).

The description of the embodiment is finished above, but various modifications and changes can be made within the scope of the present invention. For example, the tone range designating means 9 may designate only a pitch (difference in pitch, for example, one octave) between the highest note and the lowest note. If desired, all of the degree designating means 7, the range determining means 8, and the scale designating means 9 can be automated so as to variably and automatically designate data. Instead of the random number generating means 1, an original time series generating means or an original time series sampling means for generating an original time series having an appropriate characteristic such that the gain changes depending on the frequency of fluctuation can be used. Further, a function of displaying or outputting the fluctuation frequency characteristic of the melody to be composed may be added. This is convenient when the user does not know the fluctuation frequency characteristic of the original time series. Preferably, when the degree n is designated by the degree designation means 3, the fluctuation frequency characteristic is changed to a linear value in response to the designation. It may be calculated from the frequency response characteristic of the conversion means (which is determined by the order n) and the fluctuation frequency characteristic of the original time series (this is obtained by performing spectrum analysis or the like in advance), and may be graphically displayed on a display or the like. .

[Effects of the Invention] Finally, effects and advantages of the invention described in the claims will be described.

According to the configuration described in claim (4), according to the configuration described in the claim, it is possible to generate a time series having different fluctuation frequency characteristics depending on a selected value of a variable number, while being a single linear conversion unit. Melody having various fluctuation frequency characteristics can be obtained by a composer having a simpler configuration than the conventional one.

According to claims (1) and (6), a composer and composition method having the same effects as those of claim (4) are provided. Further, the fluctuation frequency characteristic given to the melody can be freely selected from 1 / f ⁿ (where n is a variable order). 1 / f ⁿ is convenient because it also includes 1 / f fluctuation characteristics found in many excellent songs.

Claims (2) and (3) show preferred configurations of the coefficient generating means in claim (1).

Further, according to claim (5), the melody time series whose pitch feeling is natural while basically preserving the property of the time series having the fluctuation frequency characteristic approximated to the 1 / f ⁿ characteristic Can be obtained relatively easily.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of an automatic composer according to an embodiment of the present invention, FIG. 2 is a diagram showing a specific example of a coefficient generating means of FIG. 1, and FIG. 3 is a linear diagram of FIG. FIG. 4 is a diagram showing an example of the configuration of the conversion means. FIG. 4 is a diagram showing an example of the fluctuation frequency characteristic of the generation time series when the order n = 1. FIG. 5 is a fluctuation frequency of the generation time series when the order n = 2. FIG. 6 is a diagram showing an example of characteristics, FIG. 6 is an overall circuit block diagram of an electronic musical instrument having the function of the automatic music composition machine of FIG. 1, and FIG. 7 is a flowchart of automatic music composition processing by the CPU of FIG. 1 random number generating means, 2 linear converting means, 3 order specifying means, 4 coefficient generating means, 7 frequency sequence calculating means, 8 sound range determining means.

Claims (6)

(57) [Claims] 1. A composer for generating melody data based on an output of a time series having a fluctuation frequency characteristic approximated to 1 / f ⁿ (where f is a fluctuation frequency and n is an order), Random number generating means for generating a characteristic specification data corresponding to the selected value of the order n; and characteristic setting means for variably setting the characteristic specification data corresponding to the selected value of the order n. has a coefficient generating means, by linearly converting the time sequence of the random numbers in the linear transform coefficients, and a linear transformation means for generating a time series with a 1 / f ⁿ approximate frequency characteristic to order the selected value A composer characterized in that: 2. A composer according to claim 1, wherein said linear transformation coefficient generated by said coefficient generation means is given by a polynomial of said characteristic designation data. 3. The method according to claim 2, wherein the polynomial is: (Where a _r is the r-th (r ≧ 1) linear transformation coefficient,
n is given by specific designation data), and a ₀ is given by a ₀ = 1. 4. A random number generating means for generating a time series of random numbers, a variable setting means for setting a variable selection value, and a coefficient for generating a linear conversion coefficient for the random number time series in accordance with the set selection value. Generating means, by linearly converting the time series of the random numbers with the linear conversion coefficient, from among the time series having different fluctuation frequency characteristics for each value of the variable number, the fluctuation frequency characteristic for the selected value of the variable number Linear conversion means for selectively generating a time series having, and a melody generation means for generating a melody having a similar fluctuation frequency characteristic based on the time series generated by the linear conversion means, Composer. 5. A time series generating means for generating a time series having a fluctuation frequency characteristic approximated to 1 / f ⁿ (where f is a fluctuation frequency and n is an order), and a melody to be generated is variably set in a sound range. A range composer, comprising: a range setting unit that performs the conversion; and a time series range conversion unit that converts the generated time series into a time series having the set range. 6. A composition method for generating melody data based on a time-series output having a fluctuation frequency characteristic approximated by 1 / f ⁿ (where f is a fluctuation frequency and n is an order), wherein an electronic random number is used. A time series of random numbers is generated by using a generation unit, the characteristic specification data corresponding to the selected value of the order n is variably set by an input unit, and a linear transformation coefficient for the time series of the random numbers is set by a coefficient generation unit. Generated according to the specified specified data set, linearly transform the time series of the random numbers with the linear transformation coefficient using linear transformation means, and obtain a 1 / f ⁿ approximate fluctuation frequency characteristic having the selected value as an order. A method of composing, comprising the step of generating a time series having the same.